Research on Flexible Robust Software Systems

1
Research on Flexible Robust Software Systems

• Lecture 1 Introduction

2
Position Yourself
Your Interest
Your Strength
Societal Needs

• right thing
• right time
• right place

3
Which Way to Go, Please?

• Now you have positioned yourself and selected an
  area. The easiest thing to start research is to
  read a set of papers and do incremental
  improvements.
• Extend a result to make it more general
• improve the efficiency of an algorithm and a
  protocol
• Solve a problem differently and show
  improvements.
• The real question is along which direction you
  would extend and improve? How do you know that
  you are creating a theorem or mere a homework
  problem, so to speak?

4
Von Neumanns Advice

• as a mathematical discipline travels far from
  its empirical source, or still more, if it is a
  second and third generation only indirectly
  inspired by the ideas coming from reality, it
  is beset with very grave dangers. It becomes
  more and more pure aesthetics, more and more
  purely l'art pour l'art'. ... there is grave
  danger that the subject will develop along the
  line of least resistance, that the stream, so far
  from its source, will separate into a multitude
  of insignificant branches, and that the
  discipline will become a disorganized mass of
  details and complexities."
• John Von Neumann, "The Mathematician" an essay
  in "The works of Mind", Editor E. B. Heywood,
  University of Chicago Press, 1957.
• Dont do it just because it is obvious or easy
  for you.
• Do it if it leads towards the solution of a
  significant real world problem.

5
Want to be Real? Know the Real World

• Dont invent solutions and then look for
  problems. It may work but the odds are poor.
• Listen to von Neumanns reflection of his life
  long work. Talk to developers, look at incident
  reports (what you did at the beginning of this
  class) and then develop the model problems
• what are the common threads/patterns in the
  problems that bogged down development?
• Can I develop a simple model problem that
  embodies the essence of the technical
  difficulties. (Think of dining philosopher for
  concurrency control.)
• Look at the engineering practice. What has been
  overlooked. (Think of Priority Inversion.)
• Look at the literatures. What the simplifying
  assumptions used by the authors that makes their
  live easy but the solution less than practical?

6
Generalization, Specialization and Integration

• A research problem can be generalized in some
  aspects and then specialized in some other
  aspects. The GS process creates a new
  angle/focus. For example,
• Generalization there can be some general
  software patterns to support the use of diversity
  in software architectures
• Specialization
• picking a promising application domain
• real time control
• telecomm,
• binding the high level reliability concerns with
  domain specific concerns
• Not all OS primitives are equally reliable, this
  means
• Your software pattern code could be a source of
  added complexity, the trick to escape this
  dilemma is
• Integration combining two ideas together, e.g.,
  NVP and RB. (But is it a good idea? Why?)

7
Negation of Implicit Assumptions

• Why and why not? Newton implicitly assumed that
  time and space are independent, Einstein said
  no...
• For example, both RB and NVP, outputs are checked
  by acceptance test or by voting mechanism
  respectively. It is implicitly assumed that the
  system would fail if incorrect outputs are sent
  to the system.
• We can negate the assumption. It is ok to let the
  incorrect output sent out if we can contained the
  damages.
• The negation leads us to the discovery of
  forward error recovery
• In many applications such as airline
  reservation, tele-communication systems, what
  is important is the aggregate remains sound
• In feedback control applications, what is
  important is that the system remains stable such
  that Individual control errors are correctable.

8
Next
9
Presentations

• In the 1/18/99 UIUC faculty retreat,
  communication of ideas was identified again as
  one of the most important skills in research.
  
• I heard the same message in 1979, when I was a
  Master student at CMU. It has not changed in 20
  years, probably a lot longer.
• Your presentations skills will greatly affect how
  your competence is perceived by others.
• If you have doubt, consider this
• Ph.D. qualifier is entirely verbal. To a great
  extent, you will be judged by your presentation.

10
My I3 Model

• I have thought about this for many years. To me,
  a great presentation should be
• Interesting (Tell me more..)
• Informative (Add to my knowledge base...)
• Insightful (Impart a deeper understanding ...)

11
Being Interesting

• Interesting arousing ones curiosity...
• The apparent evidences should lead to the
  innocent and the murderer should be the one that
  is least suspected.
• Shas formula Build a model of your audiences
  mind-sets, and maximize the unexpected events.
  Topics should be interesting to the audience
• For intellectuals with an inquisitive mind, this
  often means
• something seemingly unimportant actually holds
  the key
• something widely believed to be true turns out to
  be not true
• new and significant ideas, results, connections
  between different areas
• counter-intuitive, unbelievable results
• something insightful (see Slides on Insightful)

12
Being Informative

• Inform give new knowledge...
• New is relatively to your audience.
• what they already know?
• what they should know after my presentation?
• what are the steps in-between?
• Why from specific to general and from concrete
  example to abstract is the preferred style
  presentation?
• Consistent with how most people learn
• Give an easy to grasp anchor first
• Easy to involve the audience and make them think
  actively

13
Being Informative - 2

• After 2 weeks, we tend to remember (Edgar Dale).
• Passive learning
• 10 of what we read
• 20 of what we hear
• 30 of what we see (pictures)
• 50 of what we hear and see
• Active learning
• 70 of what we say
• 90 of what we say and do
• That is why I try to make you think, say and do!

14
Being Informative - 3

• Can you tell me what is the fundamental
  difference between
• the notes that you take
• the presentations that you make
• Presentation is audience centered Do they
  already know this? Can they follow the argument?
  Do I need a picture? An example?...
• Tips for getting the information across
• be energetic in your presentation, maintain eye
  contacts
• pay attention to your audiences reactions and
  adjust your presentation
• get them active and get them involved whenever
  you can

15
Being Insightful

• Insight getting right to the heart of a complex
  problem, shedding lights on,,,
• explain a seeming complex and confusing problem
  in a way that is easy to understand
• unearth hidden/unstated assumptions... And
  quickly put an argument to rest
• showing things in new angles, new lights and new
  forms and gain new understandings e.g., Fourier
  Transforms,
• demonstrate subtle but important
  connections/inter-dependencies between seemingly
  unrelated subjects e.g., identify the link
  between Aspirin and heart attack

16
High Levels of Thinking

• The key to become insightful is to exercise the
  higher level (4, 5, 6) of thinking skills as
  characterized by B. S. Bloom - analysis,
  synthesis and evaluation.
• Level 6 Evaluation judge, select, decide,
  critique, verify, debate, assess
• Level 5 Synthesis create, predict, construct,
  design, imagine, improve, produce and propose
• Level 4 Analysis Classify, categorize, derive,
  model
• Level 3 Application Calculate, solve,
  determine, apply
• Level 2 Comprehension Explain, paraphrase
• Level 1 Knowledge List, recite

17
Steps in Original Research in a New Area

• Incremental research can be done by reading
  papers and improve the results. Original ground
  breaking research requires the additional step of
  problem formulation.
• 1. A general understanding of the high level
  issues
• the mainframe computer became too fast and
  wasteful to run one program at a time. We needed
  somehow to make sure concurrent programs do not
  adversely interfere with each other.
• 2. Identify concrete instances that succinctly
  characterize the problem
• read/write a shared queue
• read/write a shared table
• 3. Distill the complex application into a
  self-contain and challenging model problem and
  derive a clean solution
• the dining philosophers problem

18
High Level Research Problems

• Can you define a concrete example for the
  following high level problems, formulate your
  model problem and then solve it?
• Reliable Cooperation
• how can the reliable component protects against
  abuse from unreliable components?
• how can the reliable component take advantage of
  the service offered by unreliable components
  without getting hurt?
• Stability
• how can the simple and reliable service continue
  to provide essential services in spite of
  failures in non-essential features?
• Flexibility and Recoverability what should be
  the architecture pattern that would allow
• new features easily added to improve the system
  service
• the failed features should be easily fixed or
  replace
• Others ...

19
Research Problems in Reliable Cooperation

• Formulate a model problem
• Identify a high level key problem the
  interaction between reliable software component
  and unreliable software components.
• Get a concrete example interaction between a
  reliable financial transaction service and your
  flashy website interface.
• Formulate a model problem Tunneling through
  unreliable component. That is, the financial
  transaction gets reliable data through unreliable
  web interface. To make the tunneling problem it
  a model problem, we need a few more application
  examples
• Example 2___________________________
• Example 3___________________________

20
Research Problems in Reliable Cooperation

• Devise a research approach
• Formulate a reasonable Interaction requirement,
  for example, in spite of the faults in the web
  interface, the financial transaction will either
  function correctly or not available to users.
  Can you formulate alternative requirements?
• Digging the tunnel Ok, we cannot make the entire
  flashy web site highly reliable. But can we make
  the data entry portion has a high integrity. That
  is,
• when it works, it will get the data correctly and
  confirmed by the user
• should the data be corrupted by the rest of the
  web site software, the corruption of the data can
  be detected and rejected.

21
Research Problems in Reliable Cooperation

• Summary
• Can you give concrete examples involve the
  interaction between reliable and unreliable
  components?
• Can you create a model problem of reliable
  cooperation between reliable and unreliable
  components?
• Can you specify the fault model of the unreliable
  components (what can go wrong and how)
• Can you develop a reliable cooperation protocol
  and prove that it can withstand the faults
  defined by the model.
• Is your fault model reasonable? That is, faults
  not addressed by the model has very low
  probability of occurrence.
• Is your protocol has reasonable efficiency? Can
  you create a convincing interactive demo?
• Challenge Can you formulate model problems for
  the stability problem and the flexibility-and-reco
  very problem?